1// Copyright 2012 the V8 project authors. All rights reserved.
2// Redistribution and use in source and binary forms, with or without
3// modification, are permitted provided that the following conditions are
4// met:
5//
6//     * Redistributions of source code must retain the above copyright
7//       notice, this list of conditions and the following disclaimer.
8//     * Redistributions in binary form must reproduce the above
9//       copyright notice, this list of conditions and the following
10//       disclaimer in the documentation and/or other materials provided
11//       with the distribution.
12//     * Neither the name of Google Inc. nor the names of its
13//       contributors may be used to endorse or promote products derived
14//       from this software without specific prior written permission.
15//
16// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27
28#include "v8.h"
29
30#include "runtime-profiler.h"
31
32#include "assembler.h"
33#include "bootstrapper.h"
34#include "code-stubs.h"
35#include "compilation-cache.h"
36#include "execution.h"
37#include "full-codegen.h"
38#include "global-handles.h"
39#include "isolate-inl.h"
40#include "mark-compact.h"
41#include "platform.h"
42#include "scopeinfo.h"
43
44namespace v8 {
45namespace internal {
46
47
48// Optimization sampler constants.
49static const int kSamplerFrameCount = 2;
50
51// Constants for statistical profiler.
52static const int kSamplerFrameWeight[kSamplerFrameCount] = { 2, 1 };
53
54static const int kSamplerTicksBetweenThresholdAdjustment = 32;
55
56static const int kSamplerThresholdInit = 3;
57static const int kSamplerThresholdMin = 1;
58static const int kSamplerThresholdDelta = 1;
59
60static const int kSamplerThresholdSizeFactorInit = 3;
61
62static const int kSizeLimit = 1500;
63
64// Constants for counter based profiler.
65
66// Number of times a function has to be seen on the stack before it is
67// optimized.
68static const int kProfilerTicksBeforeOptimization = 2;
69// If the function optimization was disabled due to high deoptimization count,
70// but the function is hot and has been seen on the stack this number of times,
71// then we try to reenable optimization for this function.
72static const int kProfilerTicksBeforeReenablingOptimization = 250;
73// If a function does not have enough type info (according to
74// FLAG_type_info_threshold), but has seen a huge number of ticks,
75// optimize it as it is.
76static const int kTicksWhenNotEnoughTypeInfo = 100;
77// We only have one byte to store the number of ticks.
78STATIC_ASSERT(kProfilerTicksBeforeOptimization < 256);
79STATIC_ASSERT(kProfilerTicksBeforeReenablingOptimization < 256);
80STATIC_ASSERT(kTicksWhenNotEnoughTypeInfo < 256);
81
82// Maximum size in bytes of generate code for a function to allow OSR.
83static const int kOSRCodeSizeAllowanceBase =
84    100 * FullCodeGenerator::kCodeSizeMultiplier;
85
86static const int kOSRCodeSizeAllowancePerTick =
87    3 * FullCodeGenerator::kCodeSizeMultiplier;
88
89// Maximum size in bytes of generated code for a function to be optimized
90// the very first time it is seen on the stack.
91static const int kMaxSizeEarlyOpt =
92    5 * FullCodeGenerator::kCodeSizeMultiplier;
93
94
95RuntimeProfiler::RuntimeProfiler(Isolate* isolate)
96    : isolate_(isolate),
97      sampler_threshold_(kSamplerThresholdInit),
98      sampler_threshold_size_factor_(kSamplerThresholdSizeFactorInit),
99      sampler_ticks_until_threshold_adjustment_(
100          kSamplerTicksBetweenThresholdAdjustment),
101      sampler_window_position_(0),
102      any_ic_changed_(false),
103      code_generated_(false) {
104  ClearSampleBuffer();
105}
106
107
108static void GetICCounts(Code* shared_code,
109                        int* ic_with_type_info_count,
110                        int* ic_total_count,
111                        int* percentage) {
112  *ic_total_count = 0;
113  *ic_with_type_info_count = 0;
114  Object* raw_info = shared_code->type_feedback_info();
115  if (raw_info->IsTypeFeedbackInfo()) {
116    TypeFeedbackInfo* info = TypeFeedbackInfo::cast(raw_info);
117    *ic_with_type_info_count = info->ic_with_type_info_count();
118    *ic_total_count = info->ic_total_count();
119  }
120  *percentage = *ic_total_count > 0
121      ? 100 * *ic_with_type_info_count / *ic_total_count
122      : 100;
123}
124
125
126void RuntimeProfiler::Optimize(JSFunction* function, const char* reason) {
127  ASSERT(function->IsOptimizable());
128
129  if (FLAG_trace_opt && function->PassesFilter(FLAG_hydrogen_filter)) {
130    PrintF("[marking ");
131    function->ShortPrint();
132    PrintF(" for recompilation, reason: %s", reason);
133    if (FLAG_type_info_threshold > 0) {
134      int typeinfo, total, percentage;
135      GetICCounts(function->shared()->code(), &typeinfo, &total, &percentage);
136      PrintF(", ICs with typeinfo: %d/%d (%d%%)", typeinfo, total, percentage);
137    }
138    PrintF("]\n");
139  }
140
141
142  if (isolate_->concurrent_recompilation_enabled() &&
143      !isolate_->bootstrapper()->IsActive()) {
144    if (isolate_->concurrent_osr_enabled() &&
145        isolate_->optimizing_compiler_thread()->IsQueuedForOSR(function)) {
146      // Do not attempt regular recompilation if we already queued this for OSR.
147      // TODO(yangguo): This is necessary so that we don't install optimized
148      // code on a function that is already optimized, since OSR and regular
149      // recompilation race.  This goes away as soon as OSR becomes one-shot.
150      return;
151    }
152    ASSERT(!function->IsInRecompileQueue());
153    function->MarkForConcurrentRecompilation();
154  } else {
155    // The next call to the function will trigger optimization.
156    function->MarkForLazyRecompilation();
157  }
158}
159
160
161void RuntimeProfiler::AttemptOnStackReplacement(JSFunction* function) {
162  // See AlwaysFullCompiler (in compiler.cc) comment on why we need
163  // Debug::has_break_points().
164  if (!FLAG_use_osr ||
165      isolate_->DebuggerHasBreakPoints() ||
166      function->IsBuiltin()) {
167    return;
168  }
169
170  SharedFunctionInfo* shared = function->shared();
171  // If the code is not optimizable, don't try OSR.
172  if (!shared->code()->optimizable()) return;
173
174  // We are not prepared to do OSR for a function that already has an
175  // allocated arguments object.  The optimized code would bypass it for
176  // arguments accesses, which is unsound.  Don't try OSR.
177  if (shared->uses_arguments()) return;
178
179  // We're using on-stack replacement: patch the unoptimized code so that
180  // any back edge in any unoptimized frame will trigger on-stack
181  // replacement for that frame.
182  if (FLAG_trace_osr) {
183    PrintF("[OSR - patching back edges in ");
184    function->PrintName();
185    PrintF("]\n");
186  }
187
188  BackEdgeTable::Patch(isolate_, shared->code());
189}
190
191
192void RuntimeProfiler::ClearSampleBuffer() {
193  memset(sampler_window_, 0, sizeof(sampler_window_));
194  memset(sampler_window_weight_, 0, sizeof(sampler_window_weight_));
195}
196
197
198int RuntimeProfiler::LookupSample(JSFunction* function) {
199  int weight = 0;
200  for (int i = 0; i < kSamplerWindowSize; i++) {
201    Object* sample = sampler_window_[i];
202    if (sample != NULL) {
203      bool fits = FLAG_lookup_sample_by_shared
204          ? (function->shared() == JSFunction::cast(sample)->shared())
205          : (function == JSFunction::cast(sample));
206      if (fits) {
207        weight += sampler_window_weight_[i];
208      }
209    }
210  }
211  return weight;
212}
213
214
215void RuntimeProfiler::AddSample(JSFunction* function, int weight) {
216  ASSERT(IsPowerOf2(kSamplerWindowSize));
217  sampler_window_[sampler_window_position_] = function;
218  sampler_window_weight_[sampler_window_position_] = weight;
219  sampler_window_position_ = (sampler_window_position_ + 1) &
220      (kSamplerWindowSize - 1);
221}
222
223
224void RuntimeProfiler::OptimizeNow() {
225  HandleScope scope(isolate_);
226
227  if (isolate_->DebuggerHasBreakPoints()) return;
228
229  DisallowHeapAllocation no_gc;
230
231  // Run through the JavaScript frames and collect them. If we already
232  // have a sample of the function, we mark it for optimizations
233  // (eagerly or lazily).
234  JSFunction* samples[kSamplerFrameCount];
235  int sample_count = 0;
236  int frame_count = 0;
237  int frame_count_limit = FLAG_watch_ic_patching ? FLAG_frame_count
238                                                 : kSamplerFrameCount;
239  for (JavaScriptFrameIterator it(isolate_);
240       frame_count++ < frame_count_limit && !it.done();
241       it.Advance()) {
242    JavaScriptFrame* frame = it.frame();
243    JSFunction* function = frame->function();
244
245    if (!FLAG_watch_ic_patching) {
246      // Adjust threshold each time we have processed
247      // a certain number of ticks.
248      if (sampler_ticks_until_threshold_adjustment_ > 0) {
249        sampler_ticks_until_threshold_adjustment_--;
250        if (sampler_ticks_until_threshold_adjustment_ <= 0) {
251          // If the threshold is not already at the minimum
252          // modify and reset the ticks until next adjustment.
253          if (sampler_threshold_ > kSamplerThresholdMin) {
254            sampler_threshold_ -= kSamplerThresholdDelta;
255            sampler_ticks_until_threshold_adjustment_ =
256                kSamplerTicksBetweenThresholdAdjustment;
257          }
258        }
259      }
260    }
261
262    SharedFunctionInfo* shared = function->shared();
263    Code* shared_code = shared->code();
264
265    if (shared_code->kind() != Code::FUNCTION) continue;
266    if (function->IsInRecompileQueue()) continue;
267
268    if (FLAG_always_osr &&
269        shared_code->allow_osr_at_loop_nesting_level() == 0) {
270      // Testing mode: always try an OSR compile for every function.
271      for (int i = 0; i < Code::kMaxLoopNestingMarker; i++) {
272        // TODO(titzer): fix AttemptOnStackReplacement to avoid this dumb loop.
273        shared_code->set_allow_osr_at_loop_nesting_level(i);
274        AttemptOnStackReplacement(function);
275      }
276      // Fall through and do a normal optimized compile as well.
277    } else if (!frame->is_optimized() &&
278        (function->IsMarkedForLazyRecompilation() ||
279         function->IsMarkedForConcurrentRecompilation() ||
280         function->IsOptimized())) {
281      // Attempt OSR if we are still running unoptimized code even though the
282      // the function has long been marked or even already been optimized.
283      int ticks = shared_code->profiler_ticks();
284      int allowance = kOSRCodeSizeAllowanceBase +
285                      ticks * kOSRCodeSizeAllowancePerTick;
286      if (shared_code->CodeSize() > allowance) {
287        if (ticks < 255) shared_code->set_profiler_ticks(ticks + 1);
288      } else {
289        int nesting = shared_code->allow_osr_at_loop_nesting_level();
290        if (nesting < Code::kMaxLoopNestingMarker) {
291          int new_nesting = nesting + 1;
292          shared_code->set_allow_osr_at_loop_nesting_level(new_nesting);
293          AttemptOnStackReplacement(function);
294        }
295      }
296      continue;
297    }
298
299    // Only record top-level code on top of the execution stack and
300    // avoid optimizing excessively large scripts since top-level code
301    // will be executed only once.
302    const int kMaxToplevelSourceSize = 10 * 1024;
303    if (shared->is_toplevel() &&
304        (frame_count > 1 || shared->SourceSize() > kMaxToplevelSourceSize)) {
305      continue;
306    }
307
308    // Do not record non-optimizable functions.
309    if (shared->optimization_disabled()) {
310      if (shared->deopt_count() >= FLAG_max_opt_count) {
311        // If optimization was disabled due to many deoptimizations,
312        // then check if the function is hot and try to reenable optimization.
313        int ticks = shared_code->profiler_ticks();
314        if (ticks >= kProfilerTicksBeforeReenablingOptimization) {
315          shared_code->set_profiler_ticks(0);
316          shared->TryReenableOptimization();
317        } else {
318          shared_code->set_profiler_ticks(ticks + 1);
319        }
320      }
321      continue;
322    }
323    if (!function->IsOptimizable()) continue;
324
325    if (FLAG_watch_ic_patching) {
326      int ticks = shared_code->profiler_ticks();
327
328      if (ticks >= kProfilerTicksBeforeOptimization) {
329        int typeinfo, total, percentage;
330        GetICCounts(shared_code, &typeinfo, &total, &percentage);
331        if (percentage >= FLAG_type_info_threshold) {
332          // If this particular function hasn't had any ICs patched for enough
333          // ticks, optimize it now.
334          Optimize(function, "hot and stable");
335        } else if (ticks >= kTicksWhenNotEnoughTypeInfo) {
336          Optimize(function, "not much type info but very hot");
337        } else {
338          shared_code->set_profiler_ticks(ticks + 1);
339          if (FLAG_trace_opt_verbose) {
340            PrintF("[not yet optimizing ");
341            function->PrintName();
342            PrintF(", not enough type info: %d/%d (%d%%)]\n",
343                   typeinfo, total, percentage);
344          }
345        }
346      } else if (!any_ic_changed_ &&
347                 shared_code->instruction_size() < kMaxSizeEarlyOpt) {
348        // If no IC was patched since the last tick and this function is very
349        // small, optimistically optimize it now.
350        Optimize(function, "small function");
351      } else {
352        shared_code->set_profiler_ticks(ticks + 1);
353      }
354    } else {  // !FLAG_watch_ic_patching
355      samples[sample_count++] = function;
356
357      int function_size = function->shared()->SourceSize();
358      int threshold_size_factor = (function_size > kSizeLimit)
359          ? sampler_threshold_size_factor_
360          : 1;
361
362      int threshold = sampler_threshold_ * threshold_size_factor;
363
364      if (LookupSample(function) >= threshold) {
365        Optimize(function, "sampler window lookup");
366      }
367    }
368  }
369  if (FLAG_watch_ic_patching) {
370    any_ic_changed_ = false;
371  } else {  // !FLAG_watch_ic_patching
372    // Add the collected functions as samples. It's important not to do
373    // this as part of collecting them because this will interfere with
374    // the sample lookup in case of recursive functions.
375    for (int i = 0; i < sample_count; i++) {
376      AddSample(samples[i], kSamplerFrameWeight[i]);
377    }
378  }
379}
380
381
382void RuntimeProfiler::SetUp() {
383  if (!FLAG_watch_ic_patching) {
384    ClearSampleBuffer();
385  }
386}
387
388
389void RuntimeProfiler::Reset() {
390  if (!FLAG_watch_ic_patching) {
391    sampler_threshold_ = kSamplerThresholdInit;
392    sampler_threshold_size_factor_ = kSamplerThresholdSizeFactorInit;
393    sampler_ticks_until_threshold_adjustment_ =
394        kSamplerTicksBetweenThresholdAdjustment;
395  }
396}
397
398
399void RuntimeProfiler::TearDown() {
400  // Nothing to do.
401}
402
403
404// Update the pointers in the sampler window after a GC.
405void RuntimeProfiler::UpdateSamplesAfterScavenge() {
406  for (int i = 0; i < kSamplerWindowSize; i++) {
407    Object* function = sampler_window_[i];
408    if (function != NULL && isolate_->heap()->InNewSpace(function)) {
409      MapWord map_word = HeapObject::cast(function)->map_word();
410      if (map_word.IsForwardingAddress()) {
411        sampler_window_[i] = map_word.ToForwardingAddress();
412      } else {
413        sampler_window_[i] = NULL;
414      }
415    }
416  }
417}
418
419
420void RuntimeProfiler::RemoveDeadSamples() {
421  for (int i = 0; i < kSamplerWindowSize; i++) {
422    Object* function = sampler_window_[i];
423    if (function != NULL &&
424        !Marking::MarkBitFrom(HeapObject::cast(function)).Get()) {
425      sampler_window_[i] = NULL;
426    }
427  }
428}
429
430
431void RuntimeProfiler::UpdateSamplesAfterCompact(ObjectVisitor* visitor) {
432  for (int i = 0; i < kSamplerWindowSize; i++) {
433    visitor->VisitPointer(&sampler_window_[i]);
434  }
435}
436
437
438} }  // namespace v8::internal
439